{
  "id": 436358,
  "title": "What do two example figures in \"Description\" mean?",
  "url": "/competitions/predict-ai-model-runtime/discussion/436358",
  "author_name": "",
  "post_date": "2023-09-02T03:27:18.533185400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2F84887c9d23b297ff2cd078a30cc63583%2Flayout.png?generation=1693624903749005&amp;alt=media\" alt=\"\"></p>\n<p><strong>In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?</strong></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2Ffe16ec911d9d041b30debaee6e52179d%2Ftile.png?generation=1693625146557939&amp;alt=media\" alt=\"\"></p>\n<p><strong>In the example fused subgraph of Tile Configuration, what does kernel mean?</strong></p>",
  "messages": [
    {
      "id": "2419439",
      "postDate": "09/02/2023 03:27:18",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2F84887c9d23b297ff2cd078a30cc63583%2Flayout.png?generation=1693624903749005&amp;alt=media\" alt=\"\"></p>\n<p><strong>In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?</strong></p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2Ffe16ec911d9d041b30debaee6e52179d%2Ftile.png?generation=1693625146557939&amp;alt=media\" alt=\"\"></p>\n<p><strong>In the example fused subgraph of Tile Configuration, what does kernel mean?</strong></p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2F84887c9d23b297ff2cd078a30cc63583%2Flayout.png?generation=1693624903749005&alt=media)\n\n**In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?**\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2Ffe16ec911d9d041b30debaee6e52179d%2Ftile.png?generation=1693625146557939&alt=media)\n\n**In the example fused subgraph of Tile Configuration, what does kernel mean?**",
      "votes": null
    },
    {
      "id": "2422142",
      "postDate": "09/03/2023 18:20:59",
      "content": "<blockquote>\n  <p>In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?</p>\n</blockquote>\n<p>[2,4,6] is the output tensor shape of the add node. {1,0,2} specifies the layout of the [2,4,6] shape tensor. A layout determines the order of minor-to-major tensor dimensions. This means this tensor is laid out as [4,2,6] in the physical memory (column-major, where the first dimension are consecutive in the physical space).</p>\n<p>Likewise {0} is the layout of the output tensor of the reshape node, which is [128] shape tensor.</p>\n<blockquote>\n  <p>In the example fused subgraph of Tile Configuration, what does kernel mean?</p>\n</blockquote>\n<p>\"kernel\" refers to the convolution kernel in a convolution operation.</p>\n<p>Consider a 1D convolution computation:</p>\n<pre><code> k  ..kernel_size:\n   y[i] = x[i+k] + kernel[k]\n</code></pre>\n<p>Therefore, the \"output\" attribute in the config controls the tile size of the output y, and the \"kernel\" attribute controls the tile size of the kernel (or sometimes called filter).</p>\n<p>I hope this helps!</p>",
      "rawMarkdown": ">In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?\n\n[2,4,6] is the output tensor shape of the add node. {1,0,2} specifies the layout of the [2,4,6] shape tensor. A layout determines the order of minor-to-major tensor dimensions. This means this tensor is laid out as [4,2,6] in the physical memory (column-major, where the first dimension are consecutive in the physical space).\n\nLikewise {0} is the layout of the output tensor of the reshape node, which is [128] shape tensor.\n\n>In the example fused subgraph of Tile Configuration, what does kernel mean?\n\n\"kernel\" refers to the convolution kernel in a convolution operation.\n\nConsider a 1D convolution computation:\n```python\nfor k in 0...kernel_size:\n   y[i] = x[i+k] + kernel[k]\n```\n\nTherefore, the \"output\" attribute in the config controls the tile size of the output y, and the \"kernel\" attribute controls the tile size of the kernel (or sometimes called filter).\n\nI hope this helps!",
      "votes": null
    },
    {
      "id": "2422348",
      "postDate": "09/04/2023 00:26:03",
      "content": "<p>Thank you very much for your detailed explanation! <br>\nIt will be a great help to me.</p>",
      "rawMarkdown": "Thank you very much for your detailed explanation! \nIt will be a great help to me.",
      "votes": null
    },
    {
      "id": "2422353",
      "postDate": "09/04/2023 00:28:56",
      "content": "<p>\"{0,1,2} means what?\"</p>\n<p>you can think of {} as dim permutation, </p>\n<p>in pytorch<br>\ne.g. tensor.permute(0,1,2)</p>\n<hr>\n<p>it also means how the elements will be arranged (layout) in memory.</p>",
      "rawMarkdown": "\"{0,1,2} means what?\"\n\nyou can think of {} as dim permutation, \n\nin pytorch\ne.g. tensor.permute(0,1,2)\n\n---\n\nit also means how the elements will be arranged (layout) in memory.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2422142,
      "author_name": "mangpophothilimthana",
      "author_url": "",
      "post_date": "09/03/2023 18:20:59",
      "content": "<blockquote>\n  <p>In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?</p>\n</blockquote>\n<p>[2,4,6] is the output tensor shape of the add node. {1,0,2} specifies the layout of the [2,4,6] shape tensor. A layout determines the order of minor-to-major tensor dimensions. This means this tensor is laid out as [4,2,6] in the physical memory (column-major, where the first dimension are consecutive in the physical space).</p>\n<p>Likewise {0} is the layout of the output tensor of the reshape node, which is [128] shape tensor.</p>\n<blockquote>\n  <p>In the example fused subgraph of Tile Configuration, what does kernel mean?</p>\n</blockquote>\n<p>\"kernel\" refers to the convolution kernel in a convolution operation.</p>\n<p>Consider a 1D convolution computation:</p>\n<pre><code> k  ..kernel_size:\n   y[i] = x[i+k] + kernel[k]\n</code></pre>\n<p>Therefore, the \"output\" attribute in the config controls the tile size of the output y, and the \"kernel\" attribute controls the tile size of the kernel (or sometimes called filter).</p>\n<p>I hope this helps!</p>",
      "votes": null,
      "replies": [
        {
          "id": 2422348,
          "author_name": "bruceqdu",
          "author_url": "",
          "post_date": "09/04/2023 00:26:03",
          "content": "<p>Thank you very much for your detailed explanation! <br>\nIt will be a great help to me.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2422353,
      "author_name": "hengck23",
      "author_url": "",
      "post_date": "09/04/2023 00:28:56",
      "content": "<p>\"{0,1,2} means what?\"</p>\n<p>you can think of {} as dim permutation, </p>\n<p>in pytorch<br>\ne.g. tensor.permute(0,1,2)</p>\n<hr>\n<p>it also means how the elements will be arranged (layout) in memory.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2419439": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2F84887c9d23b297ff2cd078a30cc63583%2Flayout.png?generation=1693624903749005&alt=media)\n\n**In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?**\n\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1723891%2Ffe16ec911d9d041b30debaee6e52179d%2Ftile.png?generation=1693625146557939&alt=media)\n\n**In the example fused subgraph of Tile Configuration, what does kernel mean?**",
    "2422142": ">In the example graph of Layout Configuration, what does each element mean? For example, [2,4,6] in add node means what? {0} means what? {0,1,2} means what?\n\n[2,4,6] is the output tensor shape of the add node. {1,0,2} specifies the layout of the [2,4,6] shape tensor. A layout determines the order of minor-to-major tensor dimensions. This means this tensor is laid out as [4,2,6] in the physical memory (column-major, where the first dimension are consecutive in the physical space).\n\nLikewise {0} is the layout of the output tensor of the reshape node, which is [128] shape tensor.\n\n>In the example fused subgraph of Tile Configuration, what does kernel mean?\n\n\"kernel\" refers to the convolution kernel in a convolution operation.\n\nConsider a 1D convolution computation:\n```python\nfor k in 0...kernel_size:\n   y[i] = x[i+k] + kernel[k]\n```\n\nTherefore, the \"output\" attribute in the config controls the tile size of the output y, and the \"kernel\" attribute controls the tile size of the kernel (or sometimes called filter).\n\nI hope this helps!",
    "2422348": "Thank you very much for your detailed explanation! \nIt will be a great help to me.",
    "2422353": "\"{0,1,2} means what?\"\n\nyou can think of {} as dim permutation, \n\nin pytorch\ne.g. tensor.permute(0,1,2)\n\n---\n\nit also means how the elements will be arranged (layout) in memory."
  },
  "source": "meta"
}